CASE STUDY 02 / AI RESEARCH PLATFORM
Problem
Miner
An AI research system that turns real user complaints into ranked, inspectable startup opportunities.
- role
- Product + AI Engineering
- stack
- Reddit data · AI analysis · Demand ranking
- status
- LIVE
- outcome
- Evidence-led problem discovery
01 / THE STAKE
A promising product idea starts with a painful problem—but anecdotes alone do not prove demand.
02 / RESEARCH PIPELINE
Raw conversations.
Ranked opportunities.
- 01COLLECT
- 02EXTRACT
- 03GROUP
- 04RANK
Relevant Reddit discussions are gathered as raw evidence of recurring user problems.
AI identifies the underlying problem, affected audience, and surrounding context.
Related signals are organized into themes so repeated demand becomes visible.
Opportunities are ordered using the available evidence while retaining their source context.
03 / KEY DECISIONS
AI finds patterns.
Evidence earns trust.
Keep evidence attached
Every synthesized problem stays connected to the discussions that support it.
- TRADE-OFF
- The product must retain and present more context around each insight.
- REJECTED
- Showing AI-generated opportunity summaries without their supporting signals.
Separate extraction from ranking
Understanding a problem and deciding its relative strength are distinct stages.
- TRADE-OFF
- The pipeline has more intermediate data and evaluation points.
- REJECTED
- Asking one model response to collect, interpret, and score everything at once.
Use signals, not certainty
Rankings help prioritize research rather than claiming an idea is automatically validated.
- TRADE-OFF
- The interface must explain why an opportunity ranks where it does.
- REJECTED
- Reducing product validation to one opaque AI score.
04 / THE EXPLORER
From a feed of noise
to research direction.
The interface keeps the ranking useful by showing the dimensions behind each opportunity.
How often the problem appears
How strongly users describe the pain
Who repeatedly experiences it
Which discussions support the insight
05 / RESULT
Problem discovery became a repeatable research workflow.
Problem Miner collects scattered conversations, identifies repeated pain, and organizes the evidence into opportunities that founders can investigate further.
WHAT I’D CHANGE
Make ranking quality continuously measurable.
I’d introduce a review loop for marking useful, duplicated, or weak opportunities, then use that feedback to evaluate extraction and ranking changes over time.
Synchronization